A Productive Trio
The three newly announced models share a common philosophy: none is designed to be the smartest model available. Instead, they are optimized for the repetitive, large-scale workloads that AI agents perform in production environments.
Gemini 3.6 Flash becomes Google's new general-purpose model for everyday tasks, including programming, knowledge work, and multimodal applications. Its primary advantage is efficiency: it delivers capabilities comparable to its predecessor while operating faster and at a lower cost—an increasingly important factor for companies deploying AI agents at scale.
Gemini 3.5 Flash-Lite pushes efficiency even further for high-volume, straightforward requests where cost matters more than sophisticated reasoning. It is Google's fastest and least expensive model in the family. According to the company, despite its smaller size, it outperforms some larger models in selected programming and automation benchmarks.
Gemini 3.5 Flash Cyber is the most unusual of the three because it is not intended for the general public. Designed exclusively for cybersecurity, it powers CodeMender, Google's security-focused AI agent. Because a model capable of identifying vulnerabilities could also assist malicious actors in exploiting them, Flash Cyber will not be publicly available. Instead, it will be accessible only to governments and trusted partners through a limited pilot program.
The Flagship That Still Hasn't Arrived
Despite Google's effort to quickly fill the gap, Tuesday's announcement cannot hide its most notable omission: Gemini 3.5 Pro, the company's flagship model, remains absent.
Originally expected in May, the release slipped without any immediate public explanation. Last week, Bloomberg reported that the delay stems from the model not yet meeting Google's internal performance targets, particularly in programming—the very area where competition has become most intense.
In a statement provided to 9to5Google, Google confirmed that the model is still undergoing testing but declined to offer a revised launch date, saying only that it would be released "as soon as it's ready."
The timing could hardly be worse. While Gemini 3.5 Pro remains in testing, xAI has introduced Grok 4.5, OpenAI has unveiled three GPT-5.6 variants, and Moonshot AI has announced Kimi K3.
What the Launch Says About Google's Strategy
Tulsee Doshi, Senior Director of Product Management for the Gemini team, described the new releases in a statement to Business Insider not as a temporary compromise but as reaching a "sweet spot" between efficiency and capability for running AI agents—a formulation that suggests a broader strategic shift.
Google appears to be betting that not every workload requires a flagship reasoning model. Instead, much of the real work performed by AI agents can be handled more efficiently by fast, low-cost models, while advanced reasoning capabilities are reserved for the smaller number of tasks that genuinely require them.
Whether this positioning will satisfy developers remains uncertain. Many are waiting for something different: the advanced reasoning capabilities promised for Gemini 3.5 Pro, capabilities that—even with their efficiency—a family of Flash models cannot fully replace.
In this context, the absence of Gemini 3.5 Pro has become more noticeable than the launch of the new Flash models themselves. Developers are now comparing not only inference speed and operating costs, but also the reasoning capabilities of flagship AI models.
Google also hinted that its attention is already shifting beyond the current delay by describing the pre-training process for Gemini 4 as its "most ambitious yet." That positioning moves the conversation toward the company's next generation of models, but it still leaves unanswered the much simpler question users have been asking for months: Where is the flagship model that was promised for this summer?
